Deep learning super sampling (DLSS) is a line of technology developed by Nvidia for real-time deep learning image enhancement and upscaling (Nvidia, 2022a). More explicitly, the technology offers a higher gaming performance by mitigating the GPU’s (Graphics processing unit) task of rendering whole high-resolution frames. Instead, a neural network anticipates what the frame should look like – lessening GPU utilization and increasing frame rates (Nvidia, 2022b).
At the end of September, the company announced a new generation of graphics cards based on its proprietary Ada Lovelace architecture. The new AD102 chip is manufactured using TSMC’s (Taiwan Semiconductor Manufacturing Company) N4 production process (4 nm) compared to the previous generations, which used Samsung’s 8 nm architecture (TechPowerUp, 2022). The new process allows a transistor count of more than double the previous generation, prompting the company to advertise its newest flagship GPU as 100% faster (Nvidia, 2022c). While the raw increase in gaming performance seems closer to the 80% mark, most gamers were not impressed, considering the higher prices and power draw of cards (PC Guide, 2022). Nonetheless, the technology behind it seems to be worth its price as Nvidia solidifies its focus on AI developments.
DLSS 3 appears to be the main selling point for the new generation, promising 450% higher frame rates in certain compatible games (Nvidia, 2022d). The huge difference stems from the new generation optical flow accelerator (OFA), which has significantly improved. Nvidia’s OFA calculates an optical flow field by capturing the course and pace of pixels moving from one frame to the other, enabling the AI to reconstruct 87.5% of the total displayed pixels (Brecher, 2022; Nvidia, 2022d). Comparing it to the previous generations, Nvidia’s Applied Deep Learning Vice President mentions that it is both “faster and higher quality” (Palumbo, 2022). This makes it less likely that DLSS 3 will be compatible with previous-generation cards.
Until now, over 35 games have supported Nvidia’s new super-resolution technology; however, the company announced that many more games would receive full support by the end of the year (Matthews, 2022). It will be interesting to watch AMD’s response to their RDNA 3 GPUs launch in November, as their upscaling technology FidelityFX Super Resolution (FSR) seemed to stack up just fine in gaming benchmarks against the previous iteration of DLSS. The technological developments in PC gaming will transform the user experience, building a more immersive environment close to reality.
References:
Brecher, H. (2022). Nvidia DLSS 3.0 erstellt ganze Frames und 87,5% aller Pixel per KI für eine deutlich bessere Gaming-Leistung. Notebookcheck. https://www.notebookcheck.com/Nvidia-DLSS-3-0-erstellt-ganze-Frames-und-87-5-aller-Pixel-per-KI-fuer-eine-deutlich-bessere-Gaming-Leistung.654245.0.html
Gapo, B. (2022). AMD RDNA 3 Release Date, Price, And Specs. GPU Mag. https://www.gpumag.com/amd-rdna-3/
Matthews, D. (2022). Here are all of the games that will support Nvidia DLSS 3. Digital Trends. https://www.digitaltrends.com/computing/here-are-all-of-the-games-that-will-support-nvidia-dlss-3/
Nvidia. (2022a). Deep Learning Super Sampling (DLSS) Technology. NVIDIA.
https://www.nvidia.com/nl-nl/geforce/technologies/dlss/
Nvidia. (2022b). Artificial Neural Network. NVIDIA Developer. https://developer.nvidia.com/discover/artificial-neural-network
Nvidia. (2022c). NVIDIA Ada Lovelace Architecture. NVIDIA.
https://www.nvidia.com/nl-nl/geforce/ada-lovelace-architecture/
Nvidia. (2022d). Introducing NVIDIA DLSS 3. NVIDIA.
https://www.nvidia.com/en-us/geforce/news/dlss3-ai-powered-neural-graphics-innovations/
Palumbo, A. (2022). NVIDIA Engineer Says DLSS 3 on Older RTX GPUs Could Theoretically Happen, Teases RTX I/O News. Wccftech. https://wccftech.com/nvidia-engineer-says-dlss-3-on-older-rtx-gpus-could-theoretically-happen-teases-rtx-i-o-news/
PC Guide. (2022). RTX 4090 vs RTX 3090: How Ada stacks up to Ampere. https://www.pcguide.com/gpu/rtx-4090-vs-3090/
TechPowerUp. (2022, October 14). NVIDIA AD102 GPU Specs. TechPowerUp. https://www.techpowerup.com/gpu-specs/nvidia-ad102.g1005
